AgentSlimming: Towards Efficient and Cost-Aware Multi-Agent Systems

2026-05-09

AI Summary

This work addresses the high computational overhead and inefficiency of manual design in multi-agent systems caused by redundant communication topologies. To this end, the authors propose a plug-and-play compression framework that, for the first time, integrates neural network pruning and quantization principles into multi-agent settings. The framework employs a hybrid importance scoring mechanism to identify redundant agents and combines pruning, low-cost replacement strategies, and baseline-anchored validation rules to optimize the graph-structured workflow while preventing performance collapse. Experimental results demonstrate that the method reduces token consumption by 78.9% on average with negligible performance degradation—and even achieves accuracy gains in certain scenarios—thereby attaining a Pareto-optimal trade-off between computational cost and task quality.

Abstract

Large Language Model-based Multi-Agent Systems (MAS) have demonstrated remarkable capabilities in complex tasks. However, manually designing optimal communication topologies is labor-intensive, while automated expansion methods often result in bloated structures with redundant agents, leading to excessive token consumption. To address this problem, we introduce \textbf{AgentSlimming}, a plug-and-play compression framework for graph-structured multi-agent workflows. Motivated by pruning and quantization in neural networks, AgentSlimming compresses workflows by first estimating the importance score of each agent with a hybrid mechanism, and then removes redundant agents or replaces them with low-cost ones, where each operation is validated using a baseline-anchored acceptance rule to prevent performance collapse. Experiments show that AgentSlimming reduces average token cost by up to 78.9\% with negligible performance degradation, and sometimes even improves accuracy, achieving a strong Pareto-optimal trade-off between cost and quality. \textit{Our code is publicly available at https://github.com/CitrusYL/AgentSlimming

Problem

• Multi-Agent Systems
• Token Efficiency
• Redundant Agents
• Cost-Aware Compression
• Communication Topology

Innovation

• AgentSlimming
• multi-agent systems
• workflow compression
• importance scoring
• cost-aware optimization